Experimental Investigation of the Subjective Value of Information in Trading

119 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 Experimental Investigation of the Subjective Value of Information in Trading Sheizaf Rafaeli Graduate School of Business University of Haifa sheizaf@rafaeli.net Daphne R. Raban Graduate School of Business University of Haifa draban@univ.haifa.ac.il ABSTRACT Subjective judgments regarding information are important for the design of information systems. This study examines the endowment effect in the context of evaluating information. Theoretically, value judgments that affect the demand for information are influenced by ownership rights, a phenomenon known as the endowment effect in trading situations. In a simple computer simulated business game, 31 participants conducted a management task in which they were provided opportunities to buy or sell information. The bidding mechanism was incentive compatible. Results show that, in agreement with endowment effect theory, people value information they own much more than information they do not own. This portends undertrading in information. Therefore, the subjective value of information should be considered in the design of systems. Our findings indicate a place for the subjective value of information on the WTA/WTP ratio continuum that emerges from pertinent literature. The ratio for information is similar to that of market goods. Participants had a strong inclination to purchase but not to sell information even though the profit data suggests that the use of information had no objective benefit. This preference is attributed to risk aversion rather than to loss aversion, which is the most widely accepted explanation of the endowment effect. The subjective value of information can, therefore, be affected by system design. Keywords: Information economics, perceived value, ownership, endowment effect

Journal of the Association for Information Systems (Volume 4, 2003) 119-139 120 I. INTRODUCTION The demand for information is the raison d etre of information systems. The demand for information is determined largely by its value as perceived by consumers or users of information. Gaining a better understanding of how users perceive the value of information is of importance to system developers, content providers, and the users themselves, because such an understanding may improve the flow and usage of information. What is the value of information? How can that value be determined? How does this translate into the demand for information? Is the demand for information rational? The present research proposes to investigate the subjective value of information by combining economic and psychological theory with information systems research. We use an empirical approach to suggest possible initial answers to the questions raised. Information has several unique characteristics, which render it difficult to valuate. Information is an unusual good in many aspects production, distribution, cost, and consumption. Information is both an end-product and an instrument or input into the production of other goods, decisions, and information. It is expensive to produce and cheap to reproduce [Bates 1989; Shapiro and Varian 1999]. In fact, distribution is accomplished mainly by reproduction or copying. The same content can be distributed by different media, and the price is often derived from the medium rather than from the value delivered by the content itself. In point of fact, people consume information both by sharing and by purchasing, while most other goods are consumed via purchasing only. Information is sometimes mass produced and copied, and sometimes it is sold in exclusive packages. Examples of mass produced information include stock prices and sports results. On the other hand, information may be sold in exclusive forms such as stock tips, analyst advice, and subscription to periodicals. In the case of trading in exclusive information, ownership status can be transferred between parties. The cost of information can be direct, by paying a publisher or an intermediary, and it can be indirect. Indirect information costs include time spent not doing principal work while searching for information, using low quality information leading to further costs, and purchasing advertised products from site sponsors. The quest for the value of information is further complicated by the fact that information is an experience good, meaning that its value is revealed only after consumption [Shapiro and Varian 1999; Van Alstyne 1999]. Pre- and post-consumption values of information may differ considerably and may affect future information consumption patterns. As stated before, information is transmitted via two main routes: sharing and trading (buying and selling). While technical and social attributes are important elements of the success of information systems, the value of information transmitted by sharing or trading is believed to be a central determinant of the propensity to share and of the willingness to buy or sell information [Butler 2001; Newhagen and Rafaeli 1996; Rafaeli and LaRose 1993; Steinfield and Fulk 1988]. Research on the willingness to share information has led to the finding that a central antecedent of sharing is the perception of the locus of ownership [Constant et al. 1994; Constant et al. 1996; Jarvenpaa and Staples 2000, 2001]. The propensity to share is higher when information is perceived to be owned by the organization or jointly owned by an individual and the organization as opposed to privately owned information. In this paper, we focus on trading rather than sharing. While perceived ownership has been widely documented affecting trading of a host of private and public goods [Horowitz et al. 1999; Kahneman et al. 1990; Thaler et al. 1992], the effect of the locus of perceived ownership has not yet been explored in an information trading scenario. Specifically, prior findings suggest that ownership leads to higher valuations of goods. An owned object is valued higher than the same object lacking an assigned ownership. This effect is called the endowment effect [Thaler 1980].

121 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 The purpose of the present research is to explore the effect of ownership on the perceived, or subjective, value of information in a trading environment. If ownership is an antecedent to value formation in trading, then this will provide a common ground for information sharing and trading behavior. It will also provide a broad understanding of users behavior toward information in many situations: consumption and provision of information for free or for a fee. Furthermore, in trading situations, information systems provide a unique medium for collecting data on value perceptions, for example by bidding, which may later be translated into personalized information products with valuebased pricing. Hence the present research is believed to be of significance for academic research as well as for practitioners. We begin by explaining why the value of information is subjective, and we then describe the endowment effect and its application to the study of information. Following an outline of our research tool, which is a computerized business game, and the experimental procedure, we present findings that support the significance of ownership on the perception of the value of information. II. THE VALUE OF INFORMATION Information is costly to produce but very cheap to reproduce. It is, therefore, difficult to assess its value [Shapiro and Varian 1999]. Neither cost nor the value of the information produced are related to the quantity of the product. As information is of indirect utility in support of decisions, direct utility measurement is inappropriate [Van Alstyne 1999]. Theoretically, there are three ways to assess the value of information [Ahituv 1989; Ahituv and Neumann 1986]: normative, realistic, and subjective. While user utility should be the base for calculating the price of information, utility varies by person and circumstance. Information is an experience good, the value of which is revealed only after use [Shapiro and Varian 1999]. Realistic methods are ex post and consequently inappropriate for evaluating information content (also referred to as the inspection paradox) [Van Alstyne 1999]. Therefore, we focus on the subjective value of information. Measuring subjective value is inherently problematic due to the potential mismatch between qualitative reports and quantitative or monetary scales. Another problem relating to methodology is how to translate attitudes into behavior. If we assess value using an attitude questionnaire, how will we predict people s actual information purchasing behavior? The tradition of studying decision making under uncertainty has addressed patterns of information use and the value assigned to information. The heuristics experiments [Tversky and Kahneman 1982] as well as later studies [Kahneman and Lovallo 1993] demonstrated that people tend to ignore available information such as prior probabilities, sample size, and the like. Instead, decisions are based on other subjective methods such as representativeness, availability, and adjustment and anchoring (also known collectively as heuristics). Earlier experiments have also shown that people tend to be conservative and undervalue information available for the revision of a prior opinion [Branthwaite 1975]. A recent study [Bastardi and Shafir 1998] tested the pursuit of information for daily decisions. Participants preferred to seek information and to base their choices on (objectively) noninstrumental information. In other words, people assigned positive subjective value to objectively worthless information. Theory also suggests that people seek information because it seems the right thing to do [Feldman and March 1981], implying over-demand for information and a high subjective value. People tend to accumulate information just in case they may need it in the future, again leading to excessive demand [Van Alstyne 1999]. The theoretical tension is, therefore, between studies indicating that information is under-valued and research showing that information tends to be over-valued.

Journal of the Association for Information Systems (Volume 4, 2003) 119-139 122 While previous studies have tried to elicit the subjective value of information by using surveys or semi-structured interviews [Davies 1994; Hepworth 1998; Vlahos and Ferratt 1995], this paper utilizes experimental methods. III. VALUE MEASUREMENT Subjective value has been studied experimentally for many types of market goods (also called private goods) and nonmarket goods (also called public goods). One very interesting finding of experimental research on subjective value is the discovery of a disparity between the highest amount one is willing to pay (WTP) for a good and the lowest amount one is willing to accept (WTA) as compensation for giving up the same good. Traditional economic assumptions imply that, when income effects are eliminated, the difference between WTP and WTA should be negligible (the difference should amount to the decreasing marginal utility). Empirically, however, experiments with various types of goods have shown that the differences between WTA and WTP are often substantial. Most research in this field indicates that WTA is significantly greater than WTP. By definition, WTA and WTP values are neither normative nor realistic. Instead, they are subjective values, since they represent an individual s personal perception of an object s worth for him or herself. We apply the WTA/WTP methodology as used for various types of goods in order to investigate the subjective value of information with a view to determining what characterizes information as a good. THE WTA/WTP DISPARITY The consistent, unexpectedly large and unidirectional difference between WTA and WTP observed in relation to traditional goods and services has generated much research interest. Attempts were made to explore whether the discrepancy can be explained by economic theory or whether the difference belongs to the realm of less than- or bounded-rational choice and is rooted in psychological origins. We will briefly review some of the pertinent literature on the WTA/WTP disparity and the explanations offered by economists and psychologists. We will highlight the common denominators of these two approaches. Commonly, bidding is employed as the general experimental approach for researching the values of WTA and WTP. Participants in experiments are offered the opportunity to bid for the purchase of an item, or to state a reserve price for the sale of an item. There are many bidding mechanisms and there is no specific experimental design common to all of the experiments described below. A comprehensive methodological review detailing the types of bids used in different papers can be found in Horowitz and McConnell [2002]. Using the various bidding mechanisms, researchers have demonstrated that a significant disparity exists between the values of WTA and WTP for common market goods such as chocolates, pens, and mugs [Bateman et al. 1997; Kahneman et al. 1990], and a much larger disparity with regard to nonmarket goods such as health [Shogren et al. 1994; Thaler 1980]. Trading induced-value tickets, or tokens of known value, has not shown a WTA/WTP disparity [Irwin et al. 1998; Kahneman et al. 1990; Van Dijk and Van Knippenberg 1996]. Induced value tickets or tokens are characterized by having only pure monetary value. In the case of induced value items, the expected number of trades took place, the expected number of trades being half of all possible trades. Herein lies one of the important implications of the disparity, namely, that the existence of a significant difference between WTA and WTP leads to a reluctance to trade and results in under-

123 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 trading. This was further confirmed by trading induced-value tickets of unknown value [Irwin et al. 1998; Van Dijk and Van Knippenberg 1996] as well as lottery tickets [Bar-Hillel and Neter 1996; Knetsch and Sinden 1984], which resulted in a WTA/WTP disparity and undertrading. Interestingly, uncertainty was not the cause for the disparity observed in the mugs experiment [Kahneman et al. 1990], since the bids were made on mugs marked with clearly visible price labels. The studies mentioned here as well as dozens of others [Horowitz and McConnell 2002] reveal a continuum ranging from induced (known) value tickets, where WTA is found to equal WTP, through market goods, where the disparity exists, and on to nonmarket goods, where the disparity is largest. The WTA/WTP ratio approaches unity for induced value items, being usually about 3 for market goods, while for nonmarket goods that ratio is very large, usually about 10. THEORETICAL FOUNDATION OF THE WTA/WTP DISPARITY Explanations for the WTA/WTP disparity have come from two main schools of thought, psychology and economics. Broadly, the psychological view is that decisions under uncertainty are often irrational and based on rules of thumb, heuristics. This other-than-rational behavior is also manifested in market settings. People who are endowed with a certain good will require a significantly larger sum of money to give it up than they would be willing to pay for its purchase. Since the assignment of higher value is attributed to the ownership of the object, this phenomenon was termed the endowment effect [Thaler 1980]. A later study presented the endowment effect as part of a broader theory called the mere ownership effect [Beggan 1992]. The broader claim is that the assignment of a greater value to owned objects is not limited to trading situations. The focus of the present paper is, however, on the endowment effect in trading situations. The main explanation for the endowment effect in trading offered by psychologists [Bateman et al. 1997; Kahneman et al. 1990; Thaler 1991] is loss aversion, which is part of the broader prospect theory [Kahneman and Tversky 1979]. Prospect theory, in turn, describes the value function on the basis of observed behavior rather than on economic theory (as reflected in the utility function). The value function reveals that losses are weighted more heavily than gains since the slope of the function is steeper for losses. Giving up an owned object is viewed as a loss, while the utility of obtaining the same object is viewed as a gain. Therefore, people expect a higher price for parting with an object than the price they are willing to pay to gain the object. Some studies [Chapman 1998; Hoorens et al. 1999] have tried to distinguish between the object of a trade and the transaction as such and have found that both play a role in loss aversion. Further extensions and explanations of loss aversion in the context of the WTA/WTP disparity are described by preference reversal [Thaler and Tversky 1992], the status quo bias [Thaler, Kahneman et al. 1992], the reference-dependent model [Bateman et al. 1997; Tversky and Kahneman 1991], and the anticipation of regret [Bar-Hillel and Neter 1996; Knetsch and Sinden 1984]. We turn now to the economic explanations of the endowment effect. Most experiments on the endowment effect were conducted controlling for income effects, whereby the large WTA/WTP disparity was attributed to psychological or behavioral reasons without an economic theory explaining these findings. Hanemann [1991] developed a theory based on the substitution effect. He showed, analytically, that the WTA/WTP disparity and substitution effect are inversely related. While WTP is finite, WTA can be infinite when a product has no substitutes. WTA varies with the degree of substitutability. A market good with many optional substitutes will show equality (WTA = WTP) while a good with few substitutes will show a large disparity. According to Hanemann, the disparity can

Journal of the Association for Information Systems (Volume 4, 2003) 119-139 124 range between zero (the ratio equalling one) and infinity. Shogren et al. [1994] tested Hanemann s theory and provided empirical evidence for the substitution effect theory: experimental markets for candy did not show an endowment effect, while health-risk experimental markets showed a large WTA/WTP disparity. However, contrary to the substitution effect theory, the fact remains that many of the experiments cited above have identified a WTA/WTP divergence even for simple market goods such as pens, mugs, and chocolate. Possibly a difference in experimental methods is at the root of this difference, since different bidding mechanisms were employed, and also because the theory by Hanemann and the experiments by Shogren et al. were developed for a single-sided bid whereas other experiments used double-sided auctions. Recent economic research by Kolstad and Guzman [1999] claims that an economic model can be developed explaining the divergence between WTA and WTP as being based on the uncertainty created by a lack of information about the value of the goods traded. They suggest that this model s prediction for the WTA/WTP divergence is an alternative to behavioral explanations such as the endowment effect or other economic theories such as the substitution effect. According to this model, information use is inversely related to the WTA/WTP divergence with the price of information directly related. When the price of information is high relative to expected payoffs, little information is consumed and a large disparity in WTA/WTP is observed. When the price of information is low relative to the expected payoffs, the values of WTA and WTP approach the common value, better known as the market value. Usually, however, the price of information relative to its payoff is not known a priori so that decisions have to be made under uncertainty. A study on environmental goods by Boyce et al. [1992] revealed that the disparity between WTA and WTP for environmental goods is increased by the existence of an intrinsic value. They show that WTA includes an intrinsic value which, in the case of environmental goods, is some moral responsibility of no utility or consumer value. COMMON DENOMINATORS BETWEEN PSYCHOLOGICAL AND ECONOMIC EXPLANATIONS The main psychological explanations of the WTA/WTP disparity are loss aversion, which is based on prospect theory, and the degrees of similarity and uncertainty in the cases of induced-value tokens and lottery tickets. The main economic explanations are the substitution effect, the tradeoff between the price of information and the expected payoff, and intrinsic value. The prospect theory approach received experimental economic substantiation [Horowitz et al. 1999]. Similarity observed in psychological experiments [see Chapman 1998] is equivalent to economists explanations of the substitution effect. Psychologists also acknowledged that lack of commensurability is necessary for the endowment effect to manifest itself [Kahneman et al. 1990], again a hint for the substitution effect. The immunity of induced value tickets to the endowment effect also supports the substitution effect explanation as such tickets have perfect substitutes when their values are known. The degree of uncertainty and the amount of information provided have also been researched both by psychologists and by economists, as the studies cited above exemplify. The results in all cases show similar trends. Psychological theory proposed in order to explain the WTA/WTP disparity is based on observations of human behavior. This is in line with economic models, which in this area of research are inductive and based on experimental markets rather than on traditional economic assumptions. Overall it can be said that economic and psychological

125 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 research are moving in the same direction, thus lending support to each other. The main underlying causes of the endowment effect seem to be loss aversion and the substitution effect with their respective outgrowths. Variables that influence the endowment effect are the type of good traded (inducedvalue, market, nonmarket) and the existence and availability of substitutes, which imply the availability of information on the market. IMPLICATIONS FOR THE SUBJECTIVE VALUE OF INFORMATION People pursue information for decision making as a result of the desire to reduce the uncertainty that characterizes certain decisions. Information in this sense is not a regular consumer good; it is more like a raw material consumed in the production of other goods down the value chain. The decisions as to what kind of information will aid in reducing the uncertainty, where to look for information, and what the information is worth are in themselves made under uncertainty. One rarely knows in advance what kind of information one will find, what the quality of that information will be, and to what extent it will actually reduce uncertainty. Some a priori indication of the information may be available. This is the case with film previews and article abstracts. However, such indications are partial, may be biased, and may not provide a true representation of the full information available. All of this stems from the fact that information is an experience good, the value of which is revealed only after consumption and from a lack of access to meta-information. Research that would shed light on the value of information prior to consumption or what influences value formation will be of importance to content providers, decision makers, and information system designers. Likewise, models that help to predict market making processes pertinent to the trading of information would be welcome. The result of the WTA/WTP disparity, or of the endowment effect, is that it creates undertrading. Fewer trades take place than should have occurred under standard economic assumptions. As noted earlier, lack of information contributes to an increase in the WTA/WTP divergence and hence leads to undertrading. Conversely, abundance of information suggests an accelerated pace of trade. Information is an economic catalyst. Increasing its perceived value and the demand for it should be the objective of any market-oriented organization wishing to increase the number of trades. Naturally, this is especially true for information vendors. Since information is often a crucial component of market goods, enhancing the value of that information would enhance the overall value of the goods and diminish undertrading. Substitution effect theory should predict a large WTA/WTP disparity for information. This is due to its inherent nature as an experience good, each item of content being unique. On the other hand, the abundance of free information on the Internet and searchers inclination to seek free content suggest a low subjective value for information, producing parity between WTA and WTP. In light of this contradiction we have chosen to begin our investigation with a fundamental question about the WTA and the WTP for information in order to form a basis for further research on factors influencing these values and other issues of importance. Our research question is: Where is information found on the WTA/WTP disparity continuum? Our hypothesis is that the WTA/WTP ratio for information is at least as high as holds for market goods. In other words, the WTA/WTP ratio will be greater than unity. We predict that the value will approach three, which is the average value observed for regular market goods. We expect the ratio to go no higher than the standard set for market goods.

Journal of the Association for Information Systems (Volume 4, 2003) 119-139 126 IV. RESEARCH METHODS EXPERIMENTAL INSTRUMENT A Java-based computer simulation of an easy-to-understand business game called The Lemonade Stand was used as the experimental instrument. In this simulation, the player owns a lemonade stand and must operate it so as to maximize profits by selling to passers-by. Participants are charged with making quality, pricing, and inventory decisions. Each player is permitted to modify product quality parameters for the lemonade (amounts of sugar, lemons, and ice) and may also set the price charged per cup. Furthermore, players decide how much inventory (lemons, sugar, paper cups, ice) to purchase at each round (day), such purchases to be selected from three predefined options per raw material. The players price, quality, and inventory decisions may be affected by information about weather conditions, made available prior to each day of trading, and also by the reactions of their clients, which appear as textual bubbles above their heads during the trading day. After entering all quality and inventory decisions, the player presses the sell button at his/her leisure to start the trading day. Selling is done automatically, based on the parameters entered, and cannot be interrupted by the player. During the trading, the player sees an animated representation of the stand, clients walking by, and the textual feedback with comments from clients. Typically, a round (day) would take two minutes. After playing for several consecutive days, which constitute a cumulative game, the player is notified of the clients satisfaction and the overall net profit or loss. Players were urged to maximize their profit and received a monetary reward according to the actual profit achieved. The entire simulation was conducted in the participants native language. PARTICIPANTS A total of 31 honors program undergraduate students in two groups of, respectively, 14 and 17 participated in the experiment as part of a class requirement. One group was presented first with the selling scenario, the other started with the buying scenario. The players were seated in a computer lab with an individual computer for each player. They were not allowed to interact with each other but were allowed to ask the experimenter for clarifications. PROCEDURE The experiment was preceded by a detailed in-class presentation of the simulation along with handouts that consisted of the instructions and sample screenshots. A sizeable monetary prize equal to the accumulated profit in all games was offered to the player who achieved the highest profits. Participants were told that profits could be made in two ways: (1) by trying to optimize the inventory, lemonade quality, and price per cup depending on the weather data (if available), or (2) by trading information (selling generates direct income, while buying information can generate indirect payoffs if played wisely). The experiment began with two warm-up training games of five days each, one without, the other with, weather information. The training games were followed by six real games of three days each. The players were told that the training games would not count toward the final profit calculations, being designed only for the purpose of getting acquainted with the game. Each of the six real games began with the players being presented with an opportunity either to purchase the weather information

127 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 or to sell weather information he/she owned or could keep. For each game, participants were, therefore, shown online questions about their WTA or WTP. One bidding opportunity was presented per game, the order of appearance changing between groups. Market prices of the information trades were built into the simulation but were not known or revealed to the players. They were only told that market prices were preset for each game but would not be known to them, nor would market prices be influenced by their bids (to avoid collusion). Further, participants were told that trades would be executed at market prices if the bids they offered were acceptable. This was done to ensure incentive compatibility according to the Becker-Degroot-Marschak principle [Becker, DeGroot et al. 1964]. The Becker-Degroot-Marschak principle is known in the literature as the BDM method. In BDM, a trade takes place only if bids are compatible with current market prices. BDM is, therefore, a useful method in eliciting private values and is a popular tool in studies of the endowment effect. Thus the experiment yielded three WTA values and three WTP values of the weather information for each participant. The entire experiment lasted an hour and a half, which included the presentation, the warm-up games, and the six games with bidding. MEASURES Various measures were collected from the computerized simulation of The Lemonade Stand. Stated value for the willingness to pay for weather information was entered by the participants in an empty dialog box in response to the online question: Please state the price you are willing to pay in order to purchase weather data for the entire game period. The trade will take place only if your bid complies with the current market price. Participants were asked to reconfirm their bid (with an option to change) prior to proceeding with the game. Willingness to accept payment for weather information (in U.S. dollars) was entered by the participants in an empty dialog box following the question: Please state the price for which you will agree to sell weather data for the entire game period. After selling the information you will play without weather data. The trade will take place only if your bid conforms to the current market price. Participants were asked to reconfirm their bid (with an option to change) prior to proceeding with the game. Data recorded automatically included game profits, all quality and inventory parameters entered, use of the online help option, reputation (number of clients who came following a recommendation by satisfied clients), and popularity (derived from the number of clients who bought lemonade out of the total number of clients). ANALYSIS Paired samples t-tests were performed to compare the means of WTA and WTP within groups for each of three replications of the games. The effect of conditions order was tested by an independent samples t-test of WTA and WTP values and of the ratios between groups. T-tests were performed to place the WTA/WTP ratio on the ratio continuum described in the introduction comparing the ratio to a value of 1 (WTA=WTP) and to a ratio of 3, the ratio for market goods. A general linear model for repeated measures was employed to test for differences among the three ratios. Paired samples t-tests were performed to compare the mean profits for the warm-up games with and without information.

Journal of the Association for Information Systems (Volume 4, 2003) 119-139 128 V. RESULTS The present results reveal that the WTA/WTP ratio for weather information in the lemonade simulation is about 3 and that there was a strong preference to play with information. Results shown here represent groups of 13 and 16 participants rather than the original groups of 14 and 17 because one extreme outlier was removed from each group. The extreme outliers were larger than the other data by an order of magnitude. Table 1 lists pairs of bids given in six games during the experiment. The t-value exceeds unity in all cases, although it was not always statistically significant at the 0.05 level or better. Table 1. Paired Samples Statistics to Test Differences between WTA and WTP Within Groups Type of Bid and Repetition Mean N Std. Deviation t p WTA rep* 1 Group1 22.08 13 18.95 3.01.011 WTP rep 1 Group1 7.85 13 7.36 WTA rep 2 Group1 17.92 13 25.67 1.92.078 WTP rep 2 Group1 4.61 13 2.66 WTA rep 3 Group1 19.92 13 26.79 1.71.113 WTP rep 3 Group1 7.45 13 8.14 WTA rep 1 Group2 14.91 16 12.13 3.43.004 WTP rep 1 Group2 5.10 16 2.48 WTA rep 2 Group2 15.13 16 13.09 2.89.011 WTP rep 2 Group2 6.22 16 3.81 WTA rep 3 Group2 10.44 16 6.85 1.99.065 WTP rep 3 Group2 6.72 16 5.35 *rep = repetition Group 1 was given weather information and presented with the selling scenario first and then with the buying scenario when participants were not given weather information. Group 2 was presented with the opposite order: Buying scenario first and then the selling scenario. A between-groups independent samples t-test was performed for WTA values and also for WTP values. In all cases, statistical significance was not obtained, meaning there is no statistical difference between the bids of the two groups. The ratios were also compared between groups and no significant difference was found in any case. Accordingly, the data from both groups have been united into one set of data. Table 2 shows the mean values of the WTA/WTP ratios (mean of ratios), the t-value and its significance for a one-sample test once with the test value being one and once with the test value equaling three. The results indicate that the mean ratio is significantly different than one but not significantly different from three.

129 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 Table 2. One Sample Statistics to Test Differences between the Mean WTA/WTP Ratio (Denoted Ratio in the Table) and Values of 1 (Meaning WTA = WTP) or 3. Group Ratio N Mean Std. Deviation t (test value=1) p t (test value=3) p Ratio 1 29 3.31 2.49 5.00.000 0.668 0.509 Ratio 2 29 3.49 3.91 3.43.002 0.676 0.505 Ratio 3 29 2.70 3.46 2.64.013-0.468 0.643 The general linear model for repeated measures that was used to test for differences among the three ratios produced by each participant revealed that there was no statistical difference among the three ratios (F = 1.034, p < 0.369). Table 3 summarizes the means and medians of the ratios obtained by dividing each participant s bids for WTA and WTP. Most of the experimental papers to date have reported the mean WTA value divided by the mean WTP value [Horowitz and McConnell 2002]. Table 3 includes these data as well to preserve comparability with data previously reported. Table 3. A Comparison of Means and Medians of Ratios with the Ratios of Mean and Median WTA and WTP Values Ratio 1 Ratio 2 Ratio 3 N 29 29 29 Mean of Ratios 3.31 3.49 2.70 Std. Deviation 2.49 3.91 3.46 Ratio of Means 2.86 2.98 2.09 Median of Ratios 2.05 2.00 1.43 Ratio of Medians 2.40 2.40 1.80 In order to examine whether the use of information produced an objective benefit for the participants, we examined the two warm-up games profit data. These data with the results of a paired samples t-test are summarized in Table 4. The warm-up profit data were combined after an independent samples t-test of the profits between groups revealed no significant difference for profits with or without information. Of course, data from the warm-up rounds were only used here for the purpose of assessing the presence of objective profit inherent in the access to information. Data from these rounds were not included in subsequent analyses.

Journal of the Association for Information Systems (Volume 4, 2003) 119-139 130 Table 4. Mean Profits in the Warm-Up Games with and Without Weather Information Mean game profits (U.S. dollars) With information Without information 7.58 7.40.096.924 t p Table 5 summarizes the tendency to trade information in the buying and selling bids in all six games. A total of 68.3 percent (127 out of 186) of the games were played with information. Overall, 84 trades took place out of 186 trading opportunities (45.2 percent trading). Table 5. Percent of Trades Executed in Each Type of Bid for All Games WTA WTP Percent of trades executed 26.9% 63.4% VI. DISCUSSION As predicted by the endowment effect theory, WTA for information was consistently larger than WTP for information although statistical significance of the difference was not always achieved (Table 1). We attribute the lack of statistical significance in some of the results to relatively small numbers of participants in each condition, high variance due to the absence of bidding limits and the fact that we removed only two outliers. Despite these drawbacks, a consistent and significant trend is identified where information displays a disparity between WTA and WTP values as was found previously for many market and nonmarket goods reviewed in the introduction. Further support for the existence of a substantial disparity is found by two analyses. The first analysis is the t-tests between groups comparing separately the WTA and WTP values and the individual ratios in which no significant difference was found, meaning the reproducibility of results is good. The second analysis is that of the mean ratios (Table 2) which were found to significantly diverge from unity but not differ from the repeated research-reported level of three. This means the numerator (WTA) is three times larger than the denominator (WTP). A widely used and heard slogan is that information is (or is becoming) a commodity. Nevertheless, information still has its peculiarities. It is easier to duplicate, easy to share, and ownership of it proves more difficult to enforce, even in the cases of exclusive ownership as opposed to the possession of a copy. A weighted overall mean ratio of 3.17 obtained here for WTA/WTP is typical of regular market goods [Horowitz and McConnell 2002]. Does this imply that information is a regular market good? The relatively low WTA/WTP ratio for market goods is usually attributed in the literature to the existence of fairly good substitutes. But does information, which is an experience good, have substitutes? We believe further interpretation of the results as well as further research is called for before making conclusions about the nature of information as a good.

131 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 Each player provided three WTA bids and three WTP bids, hence three ratios per player were calculated. Our aim was to see whether the ratio could be established as a stable indicator of subjective value. If no significant differences were found among the three ratios, then the ratio could be suggested as a good predictor of subjective value. The general linear model for repeated measures was performed to test for significant differences among the three ratios. No significant difference was observed among ratios, implying that a ratio is a stable predictor of subjective value despite the large variance of the values that comprise this ratio. The stability of the disparity has been researched specifically and reported to remain constant or even increase slightly over repeated trials for market goods [Morrison 2000]. The stability of the WTA/WTP disparity has been attributed to the existence of an intrinsic value which distinguishes induced-value goods from market and nonmarket goods [Boyce et al. 1992]. The determination of the intrinsic value of information is a question for further research. In Table 3, we report several descriptive data. We report the mean of ratios because we believe it is the best reflection of the individual bids which in turn reflect the individual subjective values assigned to the weather information in the lemonade game. Reporting a ratio has several advantages over reporting individual values that comprise the ratio. First, a group of ratios is likely to have less variance than its components. Second, a ratio is unit-free and enables comparison of experiments run in different countries. Third, a ratio neutralizes the object of the experiment and becomes an index in its own right. Here the ratio is an indicator of the type of good at hand. Finally, in the present case, a ratio is the traditional way to report results of experiments testing for the endowment effect. We also report the ratio of the mean values of WTA and WTP since this is the value most often found in the literature. The problem with this type of calculation is that it does not reflect the individual ratios and that it is not usable for statistical analysis since it is a single-point calculation. A ratio of means may be the value of choice when participants in an experiment bid only on one value, either WTA or WTP, because then an individual ratio cannot be established. Interestingly, mean profit values with and without the use of weather information in the warm-up games were virtually identical (Table 4). No significant difference was found for the profits of players playing with information versus without information. It is worth noting that high subjective values were assigned to information despite the lack of objective value for the information presented. This is, of course, in line with traditional MIS findings and the general notion of information overload, as in Chervany and Dickson [1974]. The players were shown their profit data at the end of each game so they were free to decide not to purchase information if it did not prove instrumental. However, our data showed that even in the face of objective uselessness of the information, participants valued information and wanted to buy it. They exhibited a bias in favor of buying more than selling, although buying and not selling had an effect on subsequent profits (buying meant paying from one s budget while selling meant earning and enlarging one s budget). Further confirmation was received from the participants during the debriefing session at the end of the game, when many expressed their strong preference toward playing with weather information, wanting to buy it, and preferring not to sell it. This is in agreement with the observations of Bastardi and Shafir [1998] and of Grant et al. [1998], who found that preference for information does not imply expected utility. It could be argued that participants may act as information dealers wanting to maximize profits by buying and selling information at every opportunity. To exhibit such behavior means to execute all three WTA deals and all three WTP deals. The bias in favor of buying information is further highlighted by the exceptionally low inclination to use free information available by clicking on the help button. The help button was selected 21 times by all participants combined (out of a possible 1,240 times), although the availability of help regarding improving performance was stressed during the presentation and in the handout.

Journal of the Association for Information Systems (Volume 4, 2003) 119-139 132 The help screen contained tips on how to improve game performance. Does the reluctance to use help mean that for-fee information is valued higher and is more desirable than free information? This question awaits further research. The strong preference to purchase information is also reflected in the percent of trades which actually took place as seen in Table 5. We assume a 50 percent percentage for trades. For WTA we see major undertrading,while for WTP we observe considerable overtrading. Participants were reluctant to sell and eager to buy information for the game as also manifested in the number of games played with information (68.3 percent of the games played with information). The overall average of the trades that took place was 45.2 percent, which shows only a small trend for undertrading. These results suggest that information may be a more marketable good than many other market goods because the desire to buy information is very high. The distinct preference to own information is also reflected in the relatively high values assigned both to WTA and WTP, which resulted in a relatively low ratio. The high value of WTA implies not only the reluctance to sell but also the value assigned to non-use of information for the purpose of playing the game. The present experiment was a one-sided, open-ended bid for the purchase of information which had potential instrumental value for the lemonade game. One-sided, open-ended bidding in auctions is known to produce higher prices than auctions where the decision is to accept or reject a given price [Kagel 1995]. This is yet another demonstration of a violation of procedure invariance, which is the underlying explanation of the preference reversal phenomenon mentioned in the introduction. However, in the lemonade game, bidders were told explicitly that they are not competing with each other for the information so they had no incentive to raise their bids above their private values. Each bidding question was followed by a statement reminding participants of the BDM rule (see pages 14-15) that the trade will take place only if it is compatible with current market prices, again encouraging bidders to submit their true values. It can be argued that the relatively high bids are a result of the BDM procedure, which is incentivecompatible only for the market value. In the lemonade game, a player can enter a high bid but is assured of actually paying only the market price so the player bids high strategically out of risk aversion and his/her will to verify that the bid will achieve its goal (buying or not selling). Possibly the high bids both for WTA and for WTP for information and the moderate ratio of these values can be better explained by risk aversion rather than loss aversion, which is the most widely-accepted explanation for the endowment effect. Bidders wanted to ensure access to information and did not want to risk playing without it. The relationship between information and risk aversion has been addressed by a theoretical model [Grant et al. 1998] establishing that intrinsically information-loving people are also risk-averse. Risk aversion has also been linked empirically to lack of information [Kahneman and Lovallo 1993]. The concern about strategic bidding with the purpose of making the deal rather than learning about the market value instead of bidding the true subjective value should be addressed by further experiments controlling for this issue. A closed-end BDM-based game might better predict the true subjective values. Another explanation for the high bids is that bidders had no information about the market price and received only indirect feedback whether they succeeded in purchasing or selling information from the new game settings presented to them after the bid. Lack of information tends to produce higher bids [Kagel 1995]. This corresponds well to the theory put forth by Kolstad and Guzman [1999], who showed there is an inverse relationship between the amount of available information and value of goods traded.

133 Journal of the Association for Information Systems (Volume 4, 2003) 119-139 Bidders in our game demonstrated their desire to play using weather data by entering relatively high bids for WTP (willingness to buy information) and for WTA (willingness to keep information). The result was a significantly larger number of trades executed for WTP than for WTA and a significantly larger number of games played with weather data than without. A by-product of the high value in both bids is that the ratio between the bids is relatively small and resembles that of market goods. We propose that eliciting the subjective value of a good requires an investigation into the number of trades in each type of bid. The ratio of the number of trades that took place for each type of bid is also an important dimension of the subjective value of information. The present research gives rise to many questions. Information as content is an experience good. We neither have prior information on information nor do we have substitutes. How, then, can we explain the information economy in light of these findings? Can the endowment effect phenomenon be related to the decrease in the relative demand for information while production is increasing [Lyman and Varian 2000]? One explanation could be the problem of information overload. We cannot handle and absorb the very large amounts of information at our disposal. Why do we complain of information overload while the WTA/WTP disparity and our market behavior suggest we suffer from lack of information? Perhaps, as suggested earlier [Bastardi and Shafir 1998], we misuse information. We pay attention to the wrong information for the wrong reasons. How can we reduce uncertainty regarding information and be better consumers of information? If we value our time so dearly, as suggested by experiments [Hoorens et al. 1999; Ortona and Scacciati 1992], why do we spend so much of it looking for useless information? The findings reported here, which relate the subjective value to actual or perceived ownership, suggest the possibility of designing information systems to affect either actual or perceived ownership to increase the subjective value of information. The manner in which information is presented before, during, and after it is sold can vary greatly. Systems can either highlight or understate the degree of exclusivity and proprietary access. Our findings suggest that these choices regarding presentation may affect the value assigned to such information by current or prospective customers. This distinction is especially relevant in an age when open source, freedom of information, and right to know are popular notions. In the experiment presented here, a certain type of information was used in a specific setting. In order to generalize or differentiate various types of information and find their places on the WTA/WTP ratio continuum, further experiments are needed using different types of information in different settings. For example, it can be hypothesized that information regarding a person s health will yield a high WTA/WTP ratio, possibly similar to that of public goods. Variations of the present or other simulations may be developed for training purposes to highlight the value of information and suggest ways for more efficient utilization of it. We elicited the subjective value of information by developing a theory-based experimental procedure. The main contribution of the present research is to suggest that ownership influences the subjective value of information in trading. Future research may seek out the effect of ownership both in trading and in sharing information itself. Additionally, the present research shows the subjective value of information to be consistently high by a number of parameters, including the WTA/WTP ratio, the number of games played with information, subjective reports, etc. This experiment also indicates that the consumption of information may be less than rational. Decisions related to such consumption might be caused by risk aversion. Several experimental variations and refinements are called for in order to establish a coherent theory about the subjective value of information.